Elite Cues and Mass Non-Compliance

80th Annual Midwest Political Science Association Conference

Zachary P Dickson & Sara B Hobolt

London School of Economics

Research Questions & Motivation

  • To what extent did elites motivate non-compliant behavior during the COVID-19 Pandemic?
    • What are the effects of elite cues as a condition of being targeted geographically?
    • To what extent can former US President Trump incite criminal behavior with his words?
  • Motivation – January 6th, 2021

Research Design

  • We leverage the fact that Trump called for the “liberation” of three specific states (MN, MI & VA) on April 17, 2020
  • Data – mobility (Meta 2022) & arrests (FBI 2022)
  • Time horizon – state lockdowns

How were the messages received?

Picture Note: Topic models include all quote tweets (143,171) of Trump’s LIBERATE tweets. A detailed description of text pre-processing and modeling methods are available in Appendix A.

Did the public respond?


Picture Note: Google Trends data are normalized and scaled according to time period and geography in order to represent the relative popularity of a search term on a range between 0 and 100 (Google 2020).

Identification

  • Generalized Difference-in-Differences
    • Treatment group – Counties in states where Trump called for liberation
    • Control group – Counties in states where Trump did not call for liberation
    • Estimand = targeted cue
  • Exogeneity assumption

Results – Mobility


Table 1: Cummulative estimates: Mobility

(a) DV: Mobility
Full State Democratic Counties Republican Counties
Treatment 2.284* 1.005 2.706**
(0.906) (0.631) (0.854)
Obs. 29,064 6,132 22,932
R2 0.764 0.825 0.714
(b) DV: Stay-at-home Compliance
Full State Democratic Counties Republican Counties
-1.128* -0.660 -1.336**
(0.502) (0.438) (0.476)
29,064 6,132 22,932
0.883 0.904 0.869

Note : + p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001

Estimates are from two-way fixed effects models with county and time fixed effects. Standard errors are clustered by state and time. See Appendix D in paper for full results.

Dynamic Results – Mobility


Event Study Estimates for the effect of Trump’s calls for Liberation on Mobility in Red counties.

Note: April 17th is Day 1. Full results are presented in Appendix D

Results – Crime


Substantive effects:

  • \(e^{0.121} - 1 \approx 12.7\%\) increase in arrest rate of whites in states where Trump called for liberation
  • 483 total arrests in the treatment group
  • \(483/e^{0.121} \approx 55\) additional arrests of whites in states where Trump called for liberation
ATT: Arrest rate/100k
Estimate S.E. CI.lower CI.upper p.value
ATT.avg 0.121 0.054 0.015 0.226 0.025

Discussion & Concluding Remarks

  • Trump’s calls for liberation led to an increase in non-compliant behavior
    • The effects were concentrated in red counties (mobility) and among whites (crime)
  • Elite cues can motivate non-compliant behavior
    • We’ve seen this before (i.e. Jan 6th), but identification is challenging in observational data
    • Estimates are conservative given control groups
  • Limitations & directions for future research
    • is our case (i.e. Trump/USA) unique?

Thank you!

References

Athey, Susan, Mohsen Bayati, Nikolay Doudchenko, Guido Imbens, and Khashayar Khosravi. 2021. “Matrix Completion Methods for Causal Panel Data Models.” Journal of the American Statistical Association 116 (536): 1716–30.
Bisbee, James, and Diana Da In Lee. 2022. “Objective Facts and Elite Cues: Partisan Responses to Covid-19.” The Journal of Politics 84 (3): 1278–91.
FBI. 2022. “National Incident-Based Reporting System (NIBRS), Federal Bureau of Investigation.” https://www.fbi.gov/how-we-can-help-you/more-fbi-services-and-information/ucr/nibrs.
Google. 2020. Google Trends: Search Term: “Liberate",” April. https://trends.google.com/trends/explore.
Grossman, Guy, Soojong Kim, Jonah M Rexer, and Harsha Thirumurthy. 2020. “Political Partisanship Influences Behavioral Responses to Governors’ Recommendations for COVID-19 Prevention in the United States.” Proceedings of the National Academy of Sciences 117 (39): 24144–53.
Meta. 2022. Movement Range Maps.” https://data.humdata.org/dataset/movement-range-maps.

Robustness

  • Mobility
    • Alternative measure of mobility – Google mobility data (Appendix E)
      • Retail & recreation, and Aggregate mobility
    • Alternative estimation strategy – first-difference (Appendix F)
  • Crime
    • No effect of cues on arrest rate of other races (Appendix I)
    • No effect of cues on arrest rate of entire state population (Appendix J)
    • Alternative measurement of arrests – Two-day moving average (Appendix K)
    • Alternative modeling strategy – TWFE with state & date fixed effects (Appendix K)

Crime

  • We apply the same empirical strategy at the state level to arrests for crimes related to sentiment expressed in analysis of quote-tweets of Trump’s tweets
    • NIBRS data (FBI 2022) on arrests for Disorderly conduct; Assault (Aggravated and Simple); Destruction/Damage/Vandalism of Property
  • We use race as a crude proxy for partisanship
    • 6% of black voters and 28% of Hispanic voters supported Trump in 2016
    • 54% of whites, including 62% of white men
  • Generalized Difference-in-Differences
    • Treatment group – Arrests of whites in MN, MI and VA
    • Control group – Arrests of whites in states under local lockdowns (40 states)
    • Estimand = targeted cue
    • Matrix completion for inference (Athey et al. 2021)

Background – what do we already know?

  • Elite cues have differential effects on adherance to social distancing in Democratic and Republican counties
    • Grossman et al. (2020) show that US state governors’ were more effective at motivating social distancing behavior in Democratic-leaning counties than Republican-leaning counties
    • Bisbee and Lee (2022) show that reductive messages from President Trump play a similar role as objective information (COVID-19 cases/deaths) in influencing social distancing behavior
  • Non-compliance behavior?
  • Counterfactuals?